2 * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
3 * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 * BIOS disk device handling.
34 * Ideas and algorithms from:
36 * - NetBSD libi386/biosdisk.c
37 * - FreeBSD biosboot/disk.c
42 #include <sys/limits.h>
44 #include <machine/bootinfo.h>
47 #include <bootstrap.h>
53 #define BIOS_NUMDRIVES 0x475
54 #define BIOSDISK_SECSIZE 512
55 #define BUFSIZE (1 * BIOSDISK_SECSIZE)
57 #define DT_ATAPI 0x10 /* disk type for ATAPI floppies */
58 #define WDMAJOR 0 /* major numbers for devices we frontend for */
64 #define DEBUG(fmt, args...) printf("%s: " fmt "\n", __func__, ## args)
66 #define DEBUG(fmt, args...)
70 * List of BIOS devices, translation from disk unit number to
75 int bd_unit; /* BIOS unit number */
76 int bd_cyl; /* BIOS geometry */
80 #define BD_MODEINT13 0x0000
81 #define BD_MODEEDD1 0x0001
82 #define BD_MODEEDD3 0x0002
83 #define BD_MODEMASK 0x0003
84 #define BD_FLOPPY 0x0004
85 int bd_type; /* BIOS 'drive type' (floppy only) */
86 uint16_t bd_sectorsize; /* Sector size */
87 uint64_t bd_sectors; /* Disk size */
88 int bd_open; /* reference counter */
89 void *bd_bcache; /* buffer cache data */
91 static int nbdinfo = 0;
93 #define BD(dev) (bdinfo[(dev)->dd.d_unit])
95 static void bd_io_workaround(struct disk_devdesc *dev);
97 static int bd_read(struct disk_devdesc *dev, daddr_t dblk, int blks,
99 static int bd_write(struct disk_devdesc *dev, daddr_t dblk, int blks,
101 static int bd_int13probe(struct bdinfo *bd);
103 static int bd_init(void);
104 static int bd_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
105 char *buf, size_t *rsize);
106 static int bd_realstrategy(void *devdata, int flag, daddr_t dblk, size_t size,
107 char *buf, size_t *rsize);
108 static int bd_open(struct open_file *f, ...);
109 static int bd_close(struct open_file *f);
110 static int bd_ioctl(struct open_file *f, u_long cmd, void *data);
111 static int bd_print(int verbose);
113 struct devsw biosdisk = {
126 * Translate between BIOS device numbers and our private unit numbers.
129 bd_bios2unit(int biosdev)
133 DEBUG("looking for bios device 0x%x", biosdev);
134 for (i = 0; i < nbdinfo; i++) {
135 DEBUG("bd unit %d is BIOS device 0x%x", i, bdinfo[i].bd_unit);
136 if (bdinfo[i].bd_unit == biosdev)
143 bd_unit2bios(int unit)
146 if ((unit >= 0) && (unit < nbdinfo))
147 return (bdinfo[unit].bd_unit);
152 * Quiz the BIOS for disk devices, save a little info about them.
157 int base, unit, nfd = 0;
159 /* sequence 0, 0x80 */
160 for (base = 0; base <= 0x80; base += 0x80) {
161 for (unit = base; (nbdinfo < MAXBDDEV); unit++) {
164 * Check the BIOS equipment list for number
168 (nfd >= *(unsigned char *)PTOV(BIOS_NUMDRIVES)))
171 bdinfo[nbdinfo].bd_open = 0;
172 bdinfo[nbdinfo].bd_bcache = NULL;
173 bdinfo[nbdinfo].bd_unit = unit;
174 bdinfo[nbdinfo].bd_flags = unit < 0x80 ? BD_FLOPPY: 0;
175 if (!bd_int13probe(&bdinfo[nbdinfo]))
178 /* XXX we need "disk aliases" to make this simpler */
179 printf("BIOS drive %c: is disk%d\n", (unit < 0x80) ?
180 ('A' + unit): ('C' + unit - 0x80), nbdinfo);
186 bcache_add_dev(nbdinfo);
191 * Try to detect a device supported by the legacy int13 BIOS
194 bd_int13probe(struct bdinfo *bd)
196 struct edd_params params;
197 int ret = 1; /* assume success */
202 v86.edx = bd->bd_unit;
205 /* Don't error out if we get bad sector number, try EDD as well */
206 if (V86_CY(v86.efl) || /* carry set */
207 (v86.edx & 0xff) <= (unsigned)(bd->bd_unit & 0x7f)) /* unit # bad */
208 return (0); /* skip device */
210 if ((v86.ecx & 0x3f) == 0) /* absurd sector number */
211 ret = 0; /* set error */
213 /* Convert max cyl # -> # of cylinders */
214 bd->bd_cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
215 /* Convert max head # -> # of heads */
216 bd->bd_hds = ((v86.edx & 0xff00) >> 8) + 1;
217 bd->bd_sec = v86.ecx & 0x3f;
218 bd->bd_type = v86.ebx & 0xff;
219 bd->bd_flags |= BD_MODEINT13;
221 /* Calculate sectors count from the geometry */
222 bd->bd_sectors = bd->bd_cyl * bd->bd_hds * bd->bd_sec;
223 bd->bd_sectorsize = BIOSDISK_SECSIZE;
224 DEBUG("unit 0x%x geometry %d/%d/%d", bd->bd_unit, bd->bd_cyl,
225 bd->bd_hds, bd->bd_sec);
227 /* Determine if we can use EDD with this device. */
231 v86.edx = bd->bd_unit;
234 if (V86_CY(v86.efl) || /* carry set */
235 (v86.ebx & 0xffff) != 0xaa55 || /* signature */
236 (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
237 return (ret); /* return code from int13 AH=08 */
240 bd->bd_flags |= BD_MODEEDD1;
241 if ((v86.eax & 0xff00) >= 0x3000)
242 bd->bd_flags |= BD_MODEEDD3;
243 /* Get disk params */
244 params.len = sizeof(struct edd_params);
248 v86.edx = bd->bd_unit;
249 v86.ds = VTOPSEG(¶ms);
250 v86.esi = VTOPOFF(¶ms);
252 if (!V86_CY(v86.efl)) {
256 * Sector size must be a multiple of 512 bytes.
257 * An alternate test would be to check power of 2,
258 * powerof2(params.sector_size).
260 if (params.sector_size % BIOSDISK_SECSIZE)
261 bd->bd_sectorsize = BIOSDISK_SECSIZE;
263 bd->bd_sectorsize = params.sector_size;
265 total = bd->bd_sectorsize * params.sectors;
266 if (params.sectors != 0) {
267 /* Only update if we did not overflow. */
268 if (total > params.sectors)
269 bd->bd_sectors = params.sectors;
272 total = (uint64_t)params.cylinders *
273 params.heads * params.sectors_per_track;
274 if (total > 0 && bd->bd_sectors > total)
275 bd->bd_sectors = total;
279 DEBUG("unit 0x%x flags %x, sectors %llu, sectorsize %u",
280 bd->bd_unit, bd->bd_flags, bd->bd_sectors, bd->bd_sectorsize);
285 * Print information about disks
288 bd_print(int verbose)
290 static char line[80];
291 struct disk_devdesc dev;
297 printf("%s devices:", biosdisk.dv_name);
298 if ((ret = pager_output("\n")) != 0)
301 for (i = 0; i < nbdinfo; i++) {
302 snprintf(line, sizeof(line),
303 " disk%d: BIOS drive %c (%ju X %u):\n", i,
304 (bdinfo[i].bd_unit < 0x80) ? ('A' + bdinfo[i].bd_unit):
305 ('C' + bdinfo[i].bd_unit - 0x80),
306 (uintmax_t)bdinfo[i].bd_sectors,
307 bdinfo[i].bd_sectorsize);
308 if ((ret = pager_output(line)) != 0)
310 dev.dd.d_dev = &biosdisk;
313 dev.d_partition = -1;
315 bdinfo[i].bd_sectorsize * bdinfo[i].bd_sectors,
316 bdinfo[i].bd_sectorsize) == 0) {
317 snprintf(line, sizeof(line), " disk%d", i);
318 ret = disk_print(&dev, line, verbose);
328 * Attempt to open the disk described by (dev) for use by (f).
330 * Note that the philosophy here is "give them exactly what
331 * they ask for". This is necessary because being too "smart"
332 * about what the user might want leads to complications.
333 * (eg. given no slice or partition value, with a disk that is
334 * sliced - are they after the first BSD slice, or the DOS
338 bd_open(struct open_file *f, ...)
340 struct disk_devdesc *dev;
341 struct disk_devdesc disk;
347 dev = va_arg(ap, struct disk_devdesc *);
350 if (dev->dd.d_unit < 0 || dev->dd.d_unit >= nbdinfo)
353 if (BD(dev).bd_bcache == NULL)
354 BD(dev).bd_bcache = bcache_allocate();
357 * Read disk size from partition.
358 * This is needed to work around buggy BIOS systems returning
359 * wrong (truncated) disk media size.
360 * During bd_probe() we tested if the mulitplication of bd_sectors
361 * would overflow so it should be safe to perform here.
363 disk.dd.d_dev = dev->dd.d_dev;
364 disk.dd.d_unit = dev->dd.d_unit;
366 disk.d_partition = -1;
369 if (disk_open(&disk, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
370 BD(dev).bd_sectorsize) == 0) {
372 if (disk_ioctl(&disk, DIOCGMEDIASIZE, &size) == 0) {
373 size /= BD(dev).bd_sectorsize;
374 if (size > BD(dev).bd_sectors)
375 BD(dev).bd_sectors = size;
380 rc = disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
381 BD(dev).bd_sectorsize);
384 if (BD(dev).bd_open == 0) {
385 bcache_free(BD(dev).bd_bcache);
386 BD(dev).bd_bcache = NULL;
394 bd_close(struct open_file *f)
396 struct disk_devdesc *dev;
398 dev = (struct disk_devdesc *)f->f_devdata;
400 if (BD(dev).bd_open == 0) {
401 bcache_free(BD(dev).bd_bcache);
402 BD(dev).bd_bcache = NULL;
404 return (disk_close(dev));
408 bd_ioctl(struct open_file *f, u_long cmd, void *data)
410 struct disk_devdesc *dev;
413 dev = (struct disk_devdesc *)f->f_devdata;
415 rc = disk_ioctl(dev, cmd, data);
420 case DIOCGSECTORSIZE:
421 *(uint32_t *)data = BD(dev).bd_sectorsize;
424 *(uint64_t *)data = BD(dev).bd_sectors * BD(dev).bd_sectorsize;
433 bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
434 char *buf, size_t *rsize)
436 struct bcache_devdata bcd;
437 struct disk_devdesc *dev;
439 dev = (struct disk_devdesc *)devdata;
440 bcd.dv_strategy = bd_realstrategy;
441 bcd.dv_devdata = devdata;
442 bcd.dv_cache = BD(dev).bd_bcache;
443 return (bcache_strategy(&bcd, rw, dblk + dev->d_offset, size,
448 bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t size,
449 char *buf, size_t *rsize)
451 struct disk_devdesc *dev = (struct disk_devdesc *)devdata;
452 uint64_t disk_blocks;
454 #ifdef BD_SUPPORT_FRAGS /* XXX: sector size */
455 char fragbuf[BIOSDISK_SECSIZE];
458 fragsize = size % BIOSDISK_SECSIZE;
460 if (size % BD(dev).bd_sectorsize) {
461 panic("bd_strategy: %d bytes I/O not multiple of block size",
466 DEBUG("open_disk %p", dev);
469 * Check the value of the size argument. We do have quite small
470 * heap (64MB), but we do not know good upper limit, so we check against
471 * INT_MAX here. This will also protect us against possible overflows
472 * while translating block count to bytes.
474 if (size > INT_MAX) {
475 DEBUG("too large read: %zu bytes", size);
479 blks = size / BD(dev).bd_sectorsize;
480 if (dblk > dblk + blks)
487 * Get disk blocks, this value is either for whole disk or for
490 if (disk_ioctl(dev, DIOCGMEDIASIZE, &disk_blocks) == 0) {
491 /* DIOCGMEDIASIZE returns bytes. */
492 disk_blocks /= BD(dev).bd_sectorsize;
494 /* We should not get here. Just try to survive. */
495 disk_blocks = BD(dev).bd_sectors - dev->d_offset;
498 /* Validate source block address. */
499 if (dblk < dev->d_offset || dblk >= dev->d_offset + disk_blocks)
503 * Truncate if we are crossing disk or partition end.
505 if (dblk + blks >= dev->d_offset + disk_blocks) {
506 blks = dev->d_offset + disk_blocks - dblk;
507 size = blks * BD(dev).bd_sectorsize;
508 DEBUG("short read %d", blks);
511 switch (rw & F_MASK) {
513 DEBUG("read %d from %lld to %p", blks, dblk, buf);
515 if (blks && (rc = bd_read(dev, dblk, blks, buf))) {
516 /* Filter out floppy controller errors */
517 if (BD(dev).bd_flags != BD_FLOPPY || rc != 0x20) {
518 printf("read %d from %lld to %p, error: 0x%x\n",
519 blks, dblk, buf, rc);
523 #ifdef BD_SUPPORT_FRAGS /* XXX: sector size */
524 DEBUG("bd_strategy: frag read %d from %d+%d to %p",
525 fragsize, dblk, blks, buf + (blks * BIOSDISK_SECSIZE));
526 if (fragsize && bd_read(od, dblk + blks, 1, fragsize)) {
527 DEBUG("frag read error");
530 bcopy(fragbuf, buf + (blks * BIOSDISK_SECSIZE), fragsize);
534 DEBUG("write %d from %lld to %p", blks, dblk, buf);
536 if (blks && bd_write(dev, dblk, blks, buf)) {
537 DEBUG("write error");
540 #ifdef BD_SUPPORT_FRAGS
542 DEBUG("Attempted to write a frag");
558 bd_edd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
561 static struct edd_packet packet;
563 packet.len = sizeof(struct edd_packet);
565 packet.off = VTOPOFF(dest);
566 packet.seg = VTOPSEG(dest);
570 /* Should we Write with verify ?? 0x4302 ? */
575 v86.edx = BD(dev).bd_unit;
576 v86.ds = VTOPSEG(&packet);
577 v86.esi = VTOPOFF(&packet);
580 return (v86.eax >> 8);
585 bd_chs_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
588 uint32_t x, bpc, cyl, hd, sec;
590 bpc = BD(dev).bd_sec * BD(dev).bd_hds; /* blocks per cylinder */
592 cyl = x / bpc; /* block # / blocks per cylinder */
593 x %= bpc; /* block offset into cylinder */
594 hd = x / BD(dev).bd_sec; /* offset / blocks per track */
595 sec = x % BD(dev).bd_sec; /* offset into track */
597 /* correct sector number for 1-based BIOS numbering */
601 /* CHS doesn't support cylinders > 1023. */
608 v86.eax = 0x300 | blks;
610 v86.eax = 0x200 | blks;
611 v86.ecx = ((cyl & 0xff) << 8) | ((cyl & 0x300) >> 2) | sec;
612 v86.edx = (hd << 8) | BD(dev).bd_unit;
613 v86.es = VTOPSEG(dest);
614 v86.ebx = VTOPOFF(dest);
617 return (v86.eax >> 8);
622 bd_io_workaround(struct disk_devdesc *dev)
624 uint8_t buf[8 * 1024];
626 bd_edd_io(dev, 0xffffffff, 1, (caddr_t)buf, 0);
631 bd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
634 u_int x, sec, result, resid, retry, maxfer;
637 /* Just in case some idiot actually tries to read/write -1 blocks... */
645 * Workaround for a problem with some HP ProLiant BIOS failing to work
646 * out the boot disk after installation. hrs and kuriyama discovered
647 * this problem with an HP ProLiant DL320e Gen 8 with a 3TB HDD, and
648 * discovered that an int13h call seems to cause a buffer overrun in
649 * the bios. The problem is alleviated by doing an extra read before
650 * the buggy read. It is not immediately known whether other models
651 * are similarly affected.
653 if (dblk >= 0x100000000)
654 bd_io_workaround(dev);
656 /* Decide whether we have to bounce */
657 if (VTOP(dest) >> 20 != 0 || (BD(dev).bd_unit < 0x80 &&
658 (VTOP(dest) >> 16) !=
659 (VTOP(dest + blks * BD(dev).bd_sectorsize) >> 16))) {
662 * There is a 64k physical boundary somewhere in the
663 * destination buffer, or the destination buffer is above
664 * first 1MB of physical memory so we have to arrange a
665 * suitable bounce buffer. Allocate a buffer twice as large
666 * as we need to. Use the bottom half unless there is a break
667 * there, in which case we use the top half.
669 x = V86_IO_BUFFER_SIZE / BD(dev).bd_sectorsize;
670 x = min(x, (unsigned)blks);
671 bbuf = PTOV(V86_IO_BUFFER);
672 maxfer = x; /* limit transfers to bounce region size */
680 * Play it safe and don't cross track boundaries.
681 * (XXX this is probably unnecessary)
683 sec = dblk % BD(dev).bd_sec; /* offset into track */
684 x = min(BD(dev).bd_sec - sec, resid);
686 x = min(x, maxfer); /* fit bounce buffer */
688 /* where do we transfer to? */
689 xp = bbuf == NULL ? p : bbuf;
692 * Put your Data In, Put your Data out,
693 * Put your Data In, and shake it all about
695 if (dowrite && bbuf != NULL)
696 bcopy(p, bbuf, x * BD(dev).bd_sectorsize);
699 * Loop retrying the operation a couple of times. The BIOS
702 for (retry = 0; retry < 3; retry++) {
703 /* if retrying, reset the drive */
708 v86.edx = BD(dev).bd_unit;
712 if (BD(dev).bd_flags & BD_MODEEDD1)
713 result = bd_edd_io(dev, dblk, x, xp, dowrite);
715 result = bd_chs_io(dev, dblk, x, xp, dowrite);
721 DEBUG("Write %d sector(s) from %p (0x%x) to %lld %s", x,
722 p, VTOP(p), dblk, result ? "failed" : "ok");
724 DEBUG("Read %d sector(s) from %lld to %p (0x%x) %s", x,
725 dblk, p, VTOP(p), result ? "failed" : "ok");
729 if (!dowrite && bbuf != NULL)
730 bcopy(bbuf, p, x * BD(dev).bd_sectorsize);
731 p += (x * BD(dev).bd_sectorsize);
740 bd_read(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest)
743 return (bd_io(dev, dblk, blks, dest, 0));
747 bd_write(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest)
750 return (bd_io(dev, dblk, blks, dest, 1));
754 * Return the BIOS geometry of a given "fixed drive" in a format
755 * suitable for the legacy bootinfo structure. Since the kernel is
756 * expecting raw int 0x13/0x8 values for N_BIOS_GEOM drives, we
757 * prefer to get the information directly, rather than rely on being
758 * able to put it together from information already maintained for
759 * different purposes and for a probably different number of drives.
761 * For valid drives, the geometry is expected in the format (31..0)
762 * "000000cc cccccccc hhhhhhhh 00ssssss"; and invalid drives are
763 * indicated by returning the geometry of a "1.2M" PC-format floppy
764 * disk. And, incidentally, what is returned is not the geometry as
765 * such but the highest valid cylinder, head, and sector numbers.
768 bd_getbigeom(int bunit)
774 v86.edx = 0x80 + bunit;
778 return (((v86.ecx & 0xc0) << 18) | ((v86.ecx & 0xff00) << 8) |
779 (v86.edx & 0xff00) | (v86.ecx & 0x3f));
783 * Return a suitable dev_t value for (dev).
785 * In the case where it looks like (dev) is a SCSI disk, we allow the number of
786 * IDE disks to be specified in $num_ide_disks. There should be a Better Way.
789 bd_getdev(struct i386_devdesc *d)
791 struct disk_devdesc *dev;
798 dev = (struct disk_devdesc *)d;
799 biosdev = bd_unit2bios(dev->dd.d_unit);
800 DEBUG("unit %d BIOS device %d", dev->dd.d_unit, biosdev);
801 if (biosdev == -1) /* not a BIOS device */
803 if (disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
804 BD(dev).bd_sectorsize) != 0) /* oops, not a viable device */
809 if (biosdev < 0x80) {
810 /* floppy (or emulated floppy) or ATAPI device */
811 if (bdinfo[dev->dd.d_unit].bd_type == DT_ATAPI) {
812 /* is an ATAPI disk */
815 /* is a floppy disk */
819 /* assume an IDE disk */
822 /* default root disk unit number */
823 unit = biosdev & 0x7f;
825 /* XXX a better kludge to set the root disk unit number */
826 if ((nip = getenv("root_disk_unit")) != NULL) {
827 i = strtol(nip, &cp, 0);
828 /* check for parse error */
829 if ((cp != nip) && (*cp == 0))
833 rootdev = MAKEBOOTDEV(major, dev->d_slice + 1, unit, dev->d_partition);
834 DEBUG("dev is 0x%x\n", rootdev);